In [1]:
%matplotlib inline
from IPython.display import HTML,Image,SVG,YouTubeVideo
In [2]:
from skimage import data
import numpy as np
from skimage.morphology import disk
import skimage.filters.rank as skr
from skimage.measure import label
from skimage.morphology import watershed
from skimage.io import imread
from scipy import ndimage as ndi
import matplotlib.pyplot as plt
from skimage.segmentation import mark_boundaries
In [3]:
# segment the coins
im = data.coins()
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [4]:
# detect the eyes / nose
im = data.chelsea()
plt.imshow(im);
In [5]:
# counting the galaxies
im = data.hubble_deep_field()
plt.imshow(im);
In [6]:
im = data.page()
bg = skr.median(im, disk(10))
res = (1.*im/bg) < .8
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
plt.figure()
plt.imshow(bg,cmap=plt.cm.gray);
plt.colorbar()
plt.figure()
plt.imshow(res.astype(np.uint8),cmap=plt.cm.gray);
plt.colorbar();
In [7]:
# segment the cells
im = imread('../data/dh_phase.png')
th = im>150
th1 = im>100
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
plt.figure()
plt.imshow(1.*th+th1,cmap=plt.cm.gray)
plt.colorbar();
In [8]:
from skimage.feature import canny
ca = canny(im)
plt.figure(figsize=[10,10])
plt.imshow(ca,cmap=plt.cm.gray);
In [9]:
from skimage.morphology import watershed
from skimage.segmentation import mark_boundaries
lab,n_lab = label(th,return_num=True)
bg = th1==0
lab[bg] = n_lab+1
#med = skr.median(im,disk(5))
#gr = skr.gradient(med,disk(3))
ws = watershed(255-im,lab)
plt.imshow(mark_boundaries(im,ws))
/home/olivier/.conda/envs/py3/lib/python3.7/site-packages/skimage/morphology/_deprecated.py:5: skimage_deprecation: Function ``watershed`` is deprecated and will be removed in version 0.19. Use ``skimage.segmentation.watershed`` instead. def watershed(image, markers=None, connectivity=1, offset=None, mask=None,
Out[9]:
<matplotlib.image.AxesImage at 0x7f49b9307ad0>
In [10]:
im = imread('../data/exp0001.jpg')
plt.figure(figsize=[20,20])
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [11]:
# count red and yellow flowers
im = imread('../data/flowers.jpg')
plt.imshow(im)
plt.colorbar();
In [12]:
# find the fiber orientation
im = imread('../data/image4.png')
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [13]:
from IPython.display import YouTubeVideo
YouTubeVideo('PUcz11MLxUk', start=0, autoplay=1, theme="light", color="blue",)
Out[13]:
In [14]:
# detect stroma
im = imread('../data/Rp042826d.jpg')
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [15]:
# segment the flowers
im = imread('../data/KaneFlowers.jpg')
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [16]:
from skimage.morphology import watershed
from skimage.segmentation import mark_boundaries
gr = skr.gradient(im,disk(3))
local_min = im <= skr.minimum(im,disk(5))
lab = label(local_min)
#med = skr.median(im,disk(5))
ws = watershed(gr,lab)
plt.figure(figsize=[10,10])
plt.imshow(mark_boundaries(im,ws))
#plt.imshow(local_min)
/home/olivier/.conda/envs/py3/lib/python3.7/site-packages/skimage/morphology/_deprecated.py:5: skimage_deprecation: Function ``watershed`` is deprecated and will be removed in version 0.19. Use ``skimage.segmentation.watershed`` instead. def watershed(image, markers=None, connectivity=1, offset=None, mask=None,
Out[16]:
<matplotlib.image.AxesImage at 0x7f49b37bffd0>
In [17]:
rgb = imread('../data/4colors.JPG')
plt.figure(figsize=[20,20])
plt.imshow(rgb)
plt.colorbar();
In [18]:
r = skr.median(rgb[:,:,0],disk(1))
plt.imshow(r,cmap=plt.cm.gray)
Out[18]:
<matplotlib.image.AxesImage at 0x7f49b2ada490>
In [19]:
s = rgb.sum(axis=2)
th = s > 100
#post-processing
pth = skr.minimum(th.astype(np.uint8),disk(1))
plt.figure(figsize=[20,20])
plt.imshow(pth,cmap=plt.cm.gray)
plt.colorbar()
Out[19]:
<matplotlib.colorbar.Colorbar at 0x7f49b2cd0410>
In [20]:
lab = label(pth)
lut = np.arange(0,np.max(lab)+1)
plt.imshow(lab)
plt.colorbar()
mask = lab == 20
plt.imshow(mask)
Out[20]:
<matplotlib.image.AxesImage at 0x7f49b37c51d0>
In [21]:
from random import shuffle
shuffle(lut)
In [22]:
shuffle(lut)
plt.imshow(lut[lab])
plt.colorbar()
Out[22]:
<matplotlib.colorbar.Colorbar at 0x7f49b9379a90>
In [ ]:
In [23]:
# segment the cell
im = imread('../data/exp0001crop.jpg')
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [24]:
m = skr.median(im,disk(5))
plt.imshow(m,cmap=plt.cm.gray)
plt.colorbar()
Out[24]:
<matplotlib.colorbar.Colorbar at 0x7f49b81310d0>
In [25]:
th1 = m < 90
th2 = np.bitwise_and(110 > m,m < 130)
plt.imshow(th2)
Out[25]:
<matplotlib.image.AxesImage at 0x7f49b36c71d0>
In [26]:
markers = label(th2)
plt.imshow(markers)
plt.colorbar()
Out[26]:
<matplotlib.colorbar.Colorbar at 0x7f49b2c22b90>
In [27]:
markers[markers==3] = 2
ws = watershed(im,markers)
/home/olivier/.conda/envs/py3/lib/python3.7/site-packages/skimage/morphology/_deprecated.py:5: skimage_deprecation: Function ``watershed`` is deprecated and will be removed in version 0.19. Use ``skimage.segmentation.watershed`` instead. def watershed(image, markers=None, connectivity=1, offset=None, mask=None,
In [28]:
plt.imshow(ws)
plt.imshow(mark_boundaries(im,ws))
Out[28]:
<matplotlib.image.AxesImage at 0x7f49b2c8c3d0>
In [29]:
# segment the cell
im = imread('../data/brain.jpg')[:,:,0]
plt.figure(figsize=(10,10))
plt.imshow(im,cmap=plt.cm.gray)
plt.colorbar();
In [30]:
plt.hist(im.flatten(),255);
In [31]:
from skimage.filters import threshold_otsu
t_otsu = threshold_otsu(im)
t_otsu
Out[31]:
36
In [32]:
th = im > t_otsu
plt.figure(figsize=(10,10))
plt.imshow(th)
Out[32]:
<matplotlib.image.AxesImage at 0x7f49b285b450>
In [33]:
lab = label(th,connectivity=1)
plt.imshow(lab)
Out[33]:
<matplotlib.image.AxesImage at 0x7f49b2844f10>
In [34]:
from skimage.measure import regionprops
In [35]:
props = regionprops(lab)
brain = (lab==7).astype(np.uint8)
pp = skr.maximum(brain,disk(3))
pp = skr.minimum(pp,disk(3))
plt.imshow(pp)
Out[35]:
<matplotlib.image.AxesImage at 0x7f49b27be590>
In [36]:
for p in props:
print(p.area, p.label)
1459 1 5 2 1 3 3 4 1 5 16 6 6323 7 1 8 2 9 1 10 1 11 1 12 1 13 16 14 1 15 1 16 1 17 2 18 2 19 2 20 2 21 30 22 1 23 1 24 1 25 1 26 2 27 2 28 5 29 1 30 1 31 2 32 1 33 13 34
In [ ]:
In [ ]:
In [ ]: